Exploring Intersection of Civic Technology and Large Language Models

For years, the government has struggled to provide helpful information for residents in need. Sometimes the resources are not readily accessible, or the channel of communication is simply missing. Many people have questions about government policies but don’t know where to ask or never bother to read through official documents; instead, they end up getting incomplete or misleading information from the web. These issues call for a system that handles queries on a large scale with high efficiency and consistency - this is where large language models (LLM) come into the picture. With the help of exciting tools like LlamaIndex, we are now able to extend the knowledge base of LLMs like ChatGPT to more specific domains and tailor the model’s responses to be more relevant to the residents’ needs. This project will extend ChatGPT’s knowledge base with specific government data (such as pdfs of housing policies) and provide a frontend interface for user to interact with a ChatGPT-style chatbot. In addition to the ability to chat and ask questions, users will also be given the power to evaluate responses of the chatbot by giving a thumb up/thumb down to the response.